ingo gunther (@ingogun) header
ingo gunther (@ingogun) profile picture

All that fits the globe. Aesthetics as continuation of journalism. https://nitter.cf/t.co/MjZ1ykTJts global mapping project, Kanal X TV, also: https://nitter.cf/t.co/jmioaZ7Al3

refugee.net
Joined July 2009
Photo by ingo gunther (@ingogun)Photo by ingo gunther (@ingogun)Photo by ingo gunther (@ingogun)Photo by ingo gunther (@ingogun)Photo by ingo gunther (@ingogun)Photo by ingo gunther (@ingogun)Photo by ingo gunther (@ingogun)Photo by ingo gunther (@ingogun)
ingo gunther retweeted
Two hacker groups have released what they said were the government’s own records of its fatalities in Ukraine. Our charts show Russia’s enormous losses economist.com/graphic-detail…
3
12
43
39,477
ingo gunther retweeted
I still got it
333
2,862
118
36,880
1,232,936
ingo gunther retweeted
Another sign that the pussy cat is out of the bag. No one knows where it’s going at night. Or what it eats or catches tomorrow.
Replying to @rikiyatakehi
1/6 We find early signs that frontier LLMs are surprisingly good at *natively* reading, writing, and reasoning about neural network weights. Instead of training, models may treat raw weights like readable, editable code: diagnosing failures and patching them directly.
1
79
ingo gunther retweeted
And that was that. The end.
1
29
ingo gunther retweeted
Quite an indictment
I have a major result to share. I’ve found conclusive forensic evidence that the top 4 Chinese AI labs—DeepSeek, Moonshot, MiniMax and Ant Group—have built their newest models on NVIDIA’s most advanced Blackwell chips, which are barred from export to China. This sounds insane, but I discovered that almost every AI model carries a hidden signature of the GPU that trained it. When a model is first set up, the starting values of its weights are laid out in a pattern that depends on the exact GPU. Each combination of GPU and seed number produces a unique pattern—a sort of fingerprint of the hardware used to create the model. Traces of that pattern survive training into the published model. Again, this evidence isn’t based on any insider information or leaking. It’s based on a forensic analysis of the model weights released by the companies themselves. I searched more than six trillion combinations of chip layout and seed number and matched them to open-weights models. In doing so, I found conclusive evidence that six Chinese AI models—including DeepSeek-V4-Pro, Kimi K3, and MiniMax-M3—were initialized on NVIDIA Blackwell systems. Four of the models match Grace Blackwell, the GB200 and GB300 parts in NVIDIA’s flagship liquid-cooled rack-scale systems. I then verified these results on real hardware. Running unmodified PyTorch with the seeds I’d recovered, a Blackwell B300 reproduced DeepSeek-V4-Flash’s untouched starting weights bit for bit, and a rented GB200 reproduced Kimi K3’s to a correlation of 0.999995. A Hopper H100 reproduced none of DeepSeek-V4-Flash’s, and changing the seed by one broke the match. (DeepSeek’s seed is 953049583 by the way.) The method also was verified to match 24 known open-weights models to the hardware that developers said they used. It is exceedingly unlikely that these results are a coincidence. Across six trillion combinations, the best accidental match should reach only about 8 sigma (i.e., eight standard deviations above what random chance would produce). The real matches range from 49 to 1,385 sigma, and even the weakest single test hits 20. These were data-center-scale training runs. With some analysis of release schedules and model details, I estimate that the largest, Kimi K3 and DeepSeek-V4-Pro, used 2,000 to 3,000 Blackwell GPUs each. Clusters of this size cost on the order of $150 to $200 million each to buy, and across these labs alone would add up to more than half a billion dollars of export-controlled GPUs, not counting experimental compute or otherwise. It’s quite likely that we see multi-billion dollar expenditure on export-controlled GPUs. I am highly confident in this result, and it has been independently verified by senior researchers at OpenAI and Anthropic. You can read the full report at unemploymentlabs.com/checkpo…. There is much more I’ve identified about these models through other forms of checkpoint forensics like this, and will have more to share soon.
2
39
Quite an indictment
I have a major result to share. I’ve found conclusive forensic evidence that the top 4 Chinese AI labs—DeepSeek, Moonshot, MiniMax and Ant Group—have built their newest models on NVIDIA’s most advanced Blackwell chips, which are barred from export to China. This sounds insane, but I discovered that almost every AI model carries a hidden signature of the GPU that trained it. When a model is first set up, the starting values of its weights are laid out in a pattern that depends on the exact GPU. Each combination of GPU and seed number produces a unique pattern—a sort of fingerprint of the hardware used to create the model. Traces of that pattern survive training into the published model. Again, this evidence isn’t based on any insider information or leaking. It’s based on a forensic analysis of the model weights released by the companies themselves. I searched more than six trillion combinations of chip layout and seed number and matched them to open-weights models. In doing so, I found conclusive evidence that six Chinese AI models—including DeepSeek-V4-Pro, Kimi K3, and MiniMax-M3—were initialized on NVIDIA Blackwell systems. Four of the models match Grace Blackwell, the GB200 and GB300 parts in NVIDIA’s flagship liquid-cooled rack-scale systems. I then verified these results on real hardware. Running unmodified PyTorch with the seeds I’d recovered, a Blackwell B300 reproduced DeepSeek-V4-Flash’s untouched starting weights bit for bit, and a rented GB200 reproduced Kimi K3’s to a correlation of 0.999995. A Hopper H100 reproduced none of DeepSeek-V4-Flash’s, and changing the seed by one broke the match. (DeepSeek’s seed is 953049583 by the way.) The method also was verified to match 24 known open-weights models to the hardware that developers said they used. It is exceedingly unlikely that these results are a coincidence. Across six trillion combinations, the best accidental match should reach only about 8 sigma (i.e., eight standard deviations above what random chance would produce). The real matches range from 49 to 1,385 sigma, and even the weakest single test hits 20. These were data-center-scale training runs. With some analysis of release schedules and model details, I estimate that the largest, Kimi K3 and DeepSeek-V4-Pro, used 2,000 to 3,000 Blackwell GPUs each. Clusters of this size cost on the order of $150 to $200 million each to buy, and across these labs alone would add up to more than half a billion dollars of export-controlled GPUs, not counting experimental compute or otherwise. It’s quite likely that we see multi-billion dollar expenditure on export-controlled GPUs. I am highly confident in this result, and it has been independently verified by senior researchers at OpenAI and Anthropic. You can read the full report at unemploymentlabs.com/checkpo…. There is much more I’ve identified about these models through other forms of checkpoint forensics like this, and will have more to share soon.
2
39
And that was that. The end.
1
29
ingo gunther retweeted
🎉@ieeevis2026 has been a success so far! Abhigya and Divya shared their poster research, and Andi organized and spoke at “Molecules to Humankind.” Grateful for the thoughtful conversations, generous community, Prof. Katy Börner, and the CNS team! #DataVisualization
1
32
The United States’ self-imposed ban on commercial spying no longer makes sense, argues David Cohen. “Although economic espionage is not without risk, unilaterally renouncing the practice in the current global environment is far more dangerous.” bit.ly/4iMjVjd
5
11
4,388
ingo gunther retweeted
Ahead of US-Russia-Ukraine talks in Abu Dhabi, today's Russian papers are not expecting a breakthrough, while presenting potential discussions on European security as a “victory” for Moscow. Other papers report: ‘real wages’ falling in Russia, but underwear sales rising. #ReadingRussia
40
268
15
1,209
47,114
Americans are more likely than they were a year ago to trust information from national and local news organizations. 🗞️ Do you trust national or local news? ⬇️ #TrustInMedia #NewsTrust
5
16
1
38
3,964
ingo gunther retweeted
Replying to @mrjeffu
Reverse of “Death and the Compass” ?
1
174
Another sign that the pussy cat is out of the bag. No one knows where it’s going at night. Or what it eats or catches tomorrow.
Replying to @rikiyatakehi
1/6 We find early signs that frontier LLMs are surprisingly good at *natively* reading, writing, and reasoning about neural network weights. Instead of training, models may treat raw weights like readable, editable code: diagnosing failures and patching them directly.
1
79
ingo gunther retweeted
Replying to @NickKristof
Their killing too is unacceptable, especially when already taken hostage.
1
195
ingo gunther retweeted
This is the kind of diplomatic discourse that we need.
It's becoming nightmarish... for the Finns. Happy Halloween from the Embassy of Finland! 🎃
2
12
125
39,159
ingo gunther retweeted
Replying to @nise_yoshimi
…and stands in stark contrast to Enzensberger’s assessment of post war Germany: “The avant-garde of mediocrity”
1
55
UNGA and China vs US summed up in a few minutes
trump is playing for november. xi is playing for the next twenty years. that mismatch is why this meeting will go so well. @gzeromedia
1
46
Diplomacy
NEW from @nytimes: Xi Jinping announced at a meeting with Trump that China would send two pandas to the Atlanta zoo. Panda diplomacy — literal soft power — is a cornerstone of U.S.-China relations. Zoos pay big money to get the fur balls on loan, which draw millions of fans.
52
ingo gunther retweeted
Or can’t live with contradictions
“Authoritarianism appeals, simply, to people who cannot tolerate complexity.” — Anne Applebaum
1
2
78